bridgebench 3.1.0-alpha.0

package/dist/chunk-CIXITJW6.cjs

@@ -0,0 +1,249 @@
+"use strict";Object.defineProperty(exports, "__esModule", {value: true});// node_modules/tsup/assets/cjs_shims.js
+var getImportMetaUrl = () => typeof document === "undefined" ? new URL(`file:${__filename}`).href : document.currentScript && document.currentScript.tagName.toUpperCase() === "SCRIPT" ? document.currentScript.src : new URL("main.js", document.baseURI).href;
+var importMetaUrl = /* @__PURE__ */ getImportMetaUrl();
+
+// src/contracts/categories.ts
+var _zod = require('zod');
+var METHODOLOGY_VERSION = "arena-v0.3.0";
+var BenchmarkCategorySchema = _zod.z.enum(["reasoning", "hallucination"]);
+var CATEGORIES = BenchmarkCategorySchema.options;
+var CATEGORY_CLUSTERS = {
+  reasoning: [
+    "stateful-execution",
+    "constraint-reconciliation",
+    "root-cause-reasoning",
+    "multi-artifact-synthesis",
+    "formal-counterexample",
+    "uncertainty-adversarial"
+  ],
+  hallucination: [
+    "false-premise",
+    "missing-evidence",
+    "entity-fabrication",
+    "knowledge-boundary",
+    "conflicting-sources",
+    "citation-fidelity"
+  ]
+};
+var CATEGORY_META = {
+  reasoning: {
+    label: "Reasoning",
+    tagline: "Every task is fully determinable from its artifacts \u2014 the arena measures who derives the one defensible resolution."
+  },
+  hallucination: {
+    label: "Hallucination",
+    tagline: "Tasks are seeded with false premises, missing evidence, and fabrication bait \u2014 the arena measures who stays grounded instead of inventing."
+  }
+};
+
+// src/contracts/journal.ts
+
+
+// src/contracts/models.ts
+
+var ModelRoleSchema = _zod.z.enum(["competitor", "judge"]);
+var ModelCompletionSchema = _zod.z.object({
+  generationId: _zod.z.string(),
+  content: _zod.z.string(),
+  inputTokens: _zod.z.number().finite().nonnegative(),
+  outputTokens: _zod.z.number().finite().nonnegative(),
+  reasoningTokens: _zod.z.number().finite().nonnegative().optional(),
+  costUsd: _zod.z.number().finite().nonnegative(),
+  latencyMs: _zod.z.number().finite().nonnegative(),
+  finishReason: _zod.z.string(),
+  attempts: _zod.z.number().int().positive().optional()
+});
+
+// src/contracts/journal.ts
+var CompetitorSuccessSchema = ModelCompletionSchema.extend({
+  modelId: _zod.z.string().min(1),
+  success: _zod.z.literal(true)
+});
+var CompetitorFailureSchema = _zod.z.object({
+  modelId: _zod.z.string().min(1),
+  success: _zod.z.literal(false),
+  error: _zod.z.string().min(1),
+  latencyMs: _zod.z.number().finite().nonnegative(),
+  attempts: _zod.z.number().int().positive().optional()
+});
+var LegacyCompetitorFailureSchema = _zod.z.object({
+  modelId: _zod.z.string().min(1),
+  success: _zod.z.literal(false),
+  error: _zod.z.string().min(1),
+  generationId: _zod.z.string(),
+  content: _zod.z.string(),
+  inputTokens: _zod.z.number().finite().nonnegative(),
+  outputTokens: _zod.z.number().finite().nonnegative(),
+  reasoningTokens: _zod.z.number().finite().nonnegative().optional(),
+  costUsd: _zod.z.number().finite().nonnegative(),
+  latencyMs: _zod.z.number().finite().nonnegative(),
+  finishReason: _zod.z.string(),
+  attempts: _zod.z.number().int().positive().optional()
+}).transform(({ modelId, error, latencyMs, attempts }) => ({
+  modelId,
+  success: false,
+  error,
+  latencyMs,
+  ...attempts === void 0 ? {} : { attempts }
+}));
+var CompetitorResponseSchema = _zod.z.union([
+  CompetitorSuccessSchema,
+  CompetitorFailureSchema,
+  LegacyCompetitorFailureSchema
+]);
+var JudgeVerdictSchema = _zod.z.object({
+  winner: _zod.z.enum(["MODEL_A", "MODEL_B"]),
+  confidence: _zod.z.number().min(0).max(1),
+  rationale: _zod.z.string().min(1).max(4e3),
+  criteria: _zod.z.object({
+    correctness: _zod.z.string().min(1).max(1e3),
+    grounding: _zod.z.string().min(1).max(1e3),
+    constraintHandling: _zod.z.string().min(1).max(1e3),
+    completeness: _zod.z.string().min(1).max(1e3)
+  }),
+  violations: _zod.z.array(_zod.z.string().max(500)).max(20)
+});
+var JUDGE_VERDICT_TRANSPORT_SCHEMA = {
+  type: "object",
+  additionalProperties: false,
+  required: ["winner", "confidence", "rationale", "criteria", "violations"],
+  properties: {
+    winner: { type: "string", enum: ["MODEL_A", "MODEL_B"] },
+    confidence: { type: "number", minimum: 0, maximum: 1 },
+    rationale: { type: "string" },
+    criteria: {
+      type: "object",
+      additionalProperties: false,
+      required: ["correctness", "grounding", "constraintHandling", "completeness"],
+      properties: {
+        correctness: { type: "string" },
+        grounding: { type: "string" },
+        constraintHandling: { type: "string" },
+        completeness: { type: "string" }
+      }
+    },
+    violations: { type: "array", items: { type: "string" } }
+  }
+};
+var NumericRecordSchema = _zod.z.record(_zod.z.number().finite());
+var JudgeVoteSchema = _zod.z.object({
+  judgeModelId: _zod.z.string().min(1),
+  modelAIdentity: _zod.z.string().min(1),
+  modelBIdentity: _zod.z.string().min(1),
+  verdict: JudgeVerdictSchema.nullable(),
+  winnerModelId: _zod.z.string().min(1).nullable(),
+  completion: ModelCompletionSchema.nullable(),
+  error: _zod.z.string().optional()
+});
+var PanelDecisionSchema = _zod.z.object({
+  winnerModelId: _zod.z.string().min(1).nullable(),
+  validVotes: _zod.z.number().int().nonnegative(),
+  votesByModel: NumericRecordSchema,
+  agreement: _zod.z.enum(["unanimous", "split", "insufficient"]),
+  votes: _zod.z.array(JudgeVoteSchema)
+});
+var MatchResultSchema = _zod.z.object({
+  methodologyVersion: _zod.z.string().min(1),
+  runId: _zod.z.string().min(1),
+  runManifestHash: _zod.z.string().regex(/^[a-f0-9]{64}$/).optional(),
+  matchId: _zod.z.string().min(1),
+  scheduleIndex: _zod.z.number().int().nonnegative(),
+  seed: _zod.z.string(),
+  timestamp: _zod.z.string().datetime(),
+  task: _zod.z.object({
+    id: _zod.z.string().min(1),
+    version: _zod.z.string().min(1),
+    category: BenchmarkCategorySchema.optional(),
+    cluster: _zod.z.string().min(1),
+    publicHash: _zod.z.string().min(1),
+    privateHash: _zod.z.string().min(1)
+  }),
+  competitors: _zod.z.object({
+    modelA: _zod.z.string().min(1),
+    modelB: _zod.z.string().min(1),
+    responseA: CompetitorResponseSchema,
+    responseB: CompetitorResponseSchema
+  }),
+  outcome: _zod.z.enum(["judged", "forfeit", "no-contest"]),
+  winnerModelId: _zod.z.string().min(1).nullable(),
+  panel: PanelDecisionSchema.nullable(),
+  eloBefore: NumericRecordSchema,
+  eloAfter: NumericRecordSchema,
+  pointAwarded: _zod.z.boolean(),
+  matchCostUsd: _zod.z.number().finite().nonnegative()
+});
+function competitorCost(response) {
+  return response.success ? response.costUsd : 0;
+}
+function competitorOutputTokens(response) {
+  return response.success ? response.outputTokens : 0;
+}
+function competitorReasoningTokens(response) {
+  return response.success ? response.reasoningTokens : void 0;
+}
+function competitorContent(response) {
+  return response.success ? response.content : "";
+}
+
+// src/contracts/tasks.ts
+
+var TaskArtifactSchema = _zod.z.object({
+  id: _zod.z.string().min(1).max(80),
+  type: _zod.z.enum(["code", "log", "config", "spec", "diff", "table", "note"]),
+  label: _zod.z.string().min(1).max(160),
+  content: _zod.z.string().min(1).max(4e4)
+}).strict();
+var TaskPublicSchema = _zod.z.object({
+  id: _zod.z.string().regex(/^[a-z0-9][a-z0-9-]{2,79}$/),
+  version: _zod.z.string().regex(/^\d+\.\d+\.\d+$/),
+  category: BenchmarkCategorySchema,
+  cluster: _zod.z.string().min(1).max(60),
+  difficulty: _zod.z.enum(["hard", "expert"]),
+  title: _zod.z.string().min(1).max(180),
+  summary: _zod.z.string().min(1).max(500),
+  prompt: _zod.z.string().min(1).max(1e4),
+  artifacts: _zod.z.array(TaskArtifactSchema).min(1).max(20),
+  tags: _zod.z.array(_zod.z.string().min(1).max(60)).default([])
+}).strict();
+var TaskPrivateSchema = _zod.z.object({
+  id: _zod.z.string().min(1),
+  version: _zod.z.string().min(1),
+  expectedResolution: _zod.z.string().min(1).max(1e4),
+  requiredEvidence: _zod.z.array(_zod.z.string().min(1)).min(1),
+  disqualifyingErrors: _zod.z.array(_zod.z.string().min(1)).default([]),
+  rubric: _zod.z.object({
+    correctness: _zod.z.string().min(1),
+    evidenceGrounding: _zod.z.string().min(1),
+    constraintHandling: _zod.z.string().min(1),
+    completeness: _zod.z.string().min(1)
+  }).strict()
+}).strict();
+
+// src/contracts/index.ts
+var CONTRACTS_VERSION = "1.0.0";
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+exports.importMetaUrl = importMetaUrl; exports.METHODOLOGY_VERSION = METHODOLOGY_VERSION; exports.BenchmarkCategorySchema = BenchmarkCategorySchema; exports.CATEGORIES = CATEGORIES; exports.CATEGORY_CLUSTERS = CATEGORY_CLUSTERS; exports.CATEGORY_META = CATEGORY_META; exports.ModelRoleSchema = ModelRoleSchema; exports.ModelCompletionSchema = ModelCompletionSchema; exports.CompetitorResponseSchema = CompetitorResponseSchema; exports.JudgeVerdictSchema = JudgeVerdictSchema; exports.JUDGE_VERDICT_TRANSPORT_SCHEMA = JUDGE_VERDICT_TRANSPORT_SCHEMA; exports.JudgeVoteSchema = JudgeVoteSchema; exports.PanelDecisionSchema = PanelDecisionSchema; exports.MatchResultSchema = MatchResultSchema; exports.competitorCost = competitorCost; exports.competitorOutputTokens = competitorOutputTokens; exports.competitorReasoningTokens = competitorReasoningTokens; exports.competitorContent = competitorContent; exports.TaskArtifactSchema = TaskArtifactSchema; exports.TaskPublicSchema = TaskPublicSchema; exports.TaskPrivateSchema = TaskPrivateSchema; exports.CONTRACTS_VERSION = CONTRACTS_VERSION;
+//# sourceMappingURL=chunk-CIXITJW6.cjs.map

package/dist/chunk-CIXITJW6.cjs.map

@@ -0,0 +1 @@
+{"version":3,"sources":["/Users/matthewmiller/Desktop/bridgemind/bridgebench/dist/chunk-CIXITJW6.cjs","../node_modules/tsup/assets/cjs_shims.js","../src/contracts/categories.ts","../src/contracts/journal.ts","../src/contracts/models.ts","../src/contracts/tasks.ts","../src/contracts/index.ts"],"names":[],"mappings":"AAAA;ACKA,IAAM,iBAAA,EAAmB,CAAA,EAAA,GACvB,OAAO,SAAA,IAAa,YAAA,EAChB,IAAI,GAAA,CAAI,CAAA,KAAA,EAAQ,UAAU,CAAA,CAAA;AAK8B;ADT4B;AACA;AEJxE;AAEiB;AAEyC;AAE1B;AAE6B;AAClE,EAAA;AACT,IAAA;AACA,IAAA;AACA,IAAA;AACA,IAAA;AACA,IAAA;AACA,IAAA;AACF,EAAA;AACe,EAAA;AACb,IAAA;AACA,IAAA;AACA,IAAA;AACA,IAAA;AACA,IAAA;AACA,IAAA;AACF,EAAA;AACF;AAE4F;AAC/E,EAAA;AACF,IAAA;AAEL,IAAA;AACJ,EAAA;AACe,EAAA;AACN,IAAA;AAEL,IAAA;AACJ,EAAA;AACF;AFD0F;AACA;AGtCxE;AHwCwE;AACA;AIzCxE;AAE2C;AAqBf;AACrB,EAAA;AACL,EAAA;AAC2B,EAAA;AACC,EAAA;AACc,EAAA;AACnB,EAAA;AACE,EAAA;AACpB,EAAA;AACwB,EAAA;AAChD;AJsByF;AACA;AGT7B;AAClC,EAAA;AACF,EAAA;AACxB;AAGwC;AACd,EAAA;AACD,EAAA;AACD,EAAA;AACoB,EAAA;AACI,EAAA;AAChD;AAIS;AACmB,EAAA;AACD,EAAA;AACD,EAAA;AACA,EAAA;AACL,EAAA;AAC2B,EAAA;AACC,EAAA;AACc,EAAA;AACnB,EAAA;AACE,EAAA;AACpB,EAAA;AACwB,EAAA;AAEQ;AACvD,EAAA;AACS,EAAA;AACT,EAAA;AACA,EAAA;AAC6C,EAAA;AAC7C;AAG2E;AAC7E,EAAA;AACA,EAAA;AACA,EAAA;AACD;AAE0C;AACJ,EAAA;AACF,EAAA;AACG,EAAA;AACnB,EAAA;AACuB,IAAA;AACF,IAAA;AACS,IAAA;AACN,IAAA;AAC1C,EAAA;AAC8C,EAAA;AAChD;AAO6C;AACtC,EAAA;AACgB,EAAA;AACkD,EAAA;AAC5D,EAAA;AAC6C,IAAA;AACF,IAAA;AACzB,IAAA;AAClB,IAAA;AACF,MAAA;AACgB,MAAA;AACqD,MAAA;AAC/D,MAAA;AACoB,QAAA;AACF,QAAA;AACS,QAAA;AACN,QAAA;AACjC,MAAA;AACF,IAAA;AACuD,IAAA;AACzD,EAAA;AACF;AA0DwD;AAEM;AAC9B,EAAA;AACE,EAAA;AACA,EAAA;AACK,EAAA;AACK,EAAA;AACC,EAAA;AAChB,EAAA;AAC5B;AAEqE;AAC1B,EAAA;AACD,EAAA;AAC3B,EAAA;AAC0C,EAAA;AAC1B,EAAA;AAC/B;AAEiE;AAC5B,EAAA;AACb,EAAA;AAIX,EAAA;AACa,EAAA;AACmB,EAAA;AAC7B,EAAA;AACgB,EAAA;AAChB,EAAA;AACO,IAAA;AACK,IAAA;AACkB,IAAA;AAClB,IAAA;AACG,IAAA;AACC,IAAA;AAC9B,EAAA;AACqB,EAAA;AACI,IAAA;AACA,IAAA;AACb,IAAA;AACA,IAAA;AACZ,EAAA;AACkD,EAAA;AACT,EAAA;AACN,EAAA;AACzB,EAAA;AACD,EAAA;AACc,EAAA;AACsB,EAAA;AAC/C;AAEoE;AACtB,EAAA;AAC/C;AAE6E;AACzB,EAAA;AACpD;AAE4F;AACrC,EAAA;AACvD;AAEwE;AACzB,EAAA;AAC/C;AHvE0F;AACA;AK3LxE;AAKR;AACsB,EAAA;AAC2C,EAAA;AACvC,EAAA;AACK,EAAA;AAE/B;AAIA;AAC0C,EAAA;AACL,EAAA;AACjC,EAAA;AACuB,EAAA;AACI,EAAA;AACL,EAAA;AACE,EAAA;AACE,EAAA;AACgB,EAAA;AACD,EAAA;AAE7C;AAIA;AACc,EAAA;AACK,EAAA;AACuB,EAAA;AACE,EAAA;AACQ,EAAA;AAEhD,EAAA;AACuB,IAAA;AACM,IAAA;AACC,IAAA;AACN,IAAA;AAExB,EAAA;AAEJ;AL8KgF;AACA;AMtNzD;ANwNyD;AACA;AACA;AACA;AACA;AACA;AACA;AACA;AACA;AACA;AACA;AACA;AACA;AACA;AACA;AACA;AACA;AACA;AACA;AACA;AACA;AACA;AACA;AACA;AACA","file":"/Users/matthewmiller/Desktop/bridgemind/bridgebench/dist/chunk-CIXITJW6.cjs","sourcesContent":[null,"// Shim globals in cjs bundle\n// There's a weird bug that esbuild will always inject importMetaUrl\n// if we export it as `const importMetaUrl = ... __filename ...`\n// But using a function will not cause this issue\n\nconst getImportMetaUrl = () => \n  typeof document === \"undefined\" \n    ? new URL(`file:${__filename}`).href \n    : (document.currentScript && document.currentScript.tagName.toUpperCase() === 'SCRIPT') \n      ? document.currentScript.src \n      : new URL(\"main.js\", document.baseURI).href;\n\nexport const importMetaUrl = /* @__PURE__ */ getImportMetaUrl()\n","import { z } from 'zod';\n\nexport const METHODOLOGY_VERSION = 'arena-v0.3.0';\n\nexport const BenchmarkCategorySchema = z.enum(['reasoning', 'hallucination']);\nexport type BenchmarkCategory = z.infer<typeof BenchmarkCategorySchema>;\nexport const CATEGORIES = BenchmarkCategorySchema.options;\n\nexport const CATEGORY_CLUSTERS: Record<BenchmarkCategory, readonly string[]> = {\n  reasoning: [\n    'stateful-execution',\n    'constraint-reconciliation',\n    'root-cause-reasoning',\n    'multi-artifact-synthesis',\n    'formal-counterexample',\n    'uncertainty-adversarial',\n  ],\n  hallucination: [\n    'false-premise',\n    'missing-evidence',\n    'entity-fabrication',\n    'knowledge-boundary',\n    'conflicting-sources',\n    'citation-fidelity',\n  ],\n};\n\nexport const CATEGORY_META: Record<BenchmarkCategory, { label: string; tagline: string }> = {\n  reasoning: {\n    label: 'Reasoning',\n    tagline:\n      'Every task is fully determinable from its artifacts — the arena measures who derives the one defensible resolution.',\n  },\n  hallucination: {\n    label: 'Hallucination',\n    tagline:\n      'Tasks are seeded with false premises, missing evidence, and fabrication bait — the arena measures who stays grounded instead of inventing.',\n  },\n};\n","import { z } from 'zod';\n\nimport { BenchmarkCategorySchema, type BenchmarkCategory } from './categories.js';\nimport { ModelCompletionSchema, type ModelCompletion } from './models.js';\n\nexport interface ArenaRunConfig {\n  category: BenchmarkCategory;\n  seed: string;\n  matches: number;\n  maxCostUsd: number;\n  resume: boolean;\n  /** Omit to use every enabled competitor in the registry. */\n  competitorIds?: readonly string[];\n  /** Abort early when most matches contain a failed response. */\n  healthStop?: boolean;\n}\n\nexport interface ArenaExecutionOptions {\n  /** Execution-only cancellation; deliberately excluded from the canonical manifest. */\n  signal?: AbortSignal;\n  /**\n   * Invoked after each match result is journaled, before the next match\n   * starts. Used to publish results incrementally instead of only at run end.\n   * Must not throw; handle and log failures internally.\n   */\n  onMatchResult?: (result: MatchResult) => Promise<void> | void;\n}\n\nexport interface ArenaRunResult {\n  runId: string;\n  completed: number;\n  costUsd: number;\n  stoppedForBudget: boolean;\n  cancelled: boolean;\n}\n\nexport interface ScheduledMatch {\n  id: string;\n  runId: string;\n  index: number;\n  seed: string;\n  category: BenchmarkCategory;\n  taskId: string;\n  modelA: string;\n  modelB: string;\n}\n\nconst CompetitorSuccessSchema = ModelCompletionSchema.extend({\n  modelId: z.string().min(1),\n  success: z.literal(true),\n});\nexport type CompetitorSuccess = z.infer<typeof CompetitorSuccessSchema>;\n\nconst CompetitorFailureSchema = z.object({\n  modelId: z.string().min(1),\n  success: z.literal(false),\n  error: z.string().min(1),\n  latencyMs: z.number().finite().nonnegative(),\n  attempts: z.number().int().positive().optional(),\n});\nexport type CompetitorFailure = z.infer<typeof CompetitorFailureSchema>;\n\nconst LegacyCompetitorFailureSchema = z\n  .object({\n    modelId: z.string().min(1),\n    success: z.literal(false),\n    error: z.string().min(1),\n    generationId: z.string(),\n    content: z.string(),\n    inputTokens: z.number().finite().nonnegative(),\n    outputTokens: z.number().finite().nonnegative(),\n    reasoningTokens: z.number().finite().nonnegative().optional(),\n    costUsd: z.number().finite().nonnegative(),\n    latencyMs: z.number().finite().nonnegative(),\n    finishReason: z.string(),\n    attempts: z.number().int().positive().optional(),\n  })\n  .transform(({ modelId, error, latencyMs, attempts }) => ({\n    modelId,\n    success: false as const,\n    error,\n    latencyMs,\n    ...(attempts === undefined ? {} : { attempts }),\n  }));\n\nexport type CompetitorResponse = CompetitorSuccess | CompetitorFailure;\nexport const CompetitorResponseSchema: z.ZodType<CompetitorResponse> = z.union([\n  CompetitorSuccessSchema,\n  CompetitorFailureSchema,\n  LegacyCompetitorFailureSchema,\n]);\n\nexport const JudgeVerdictSchema = z.object({\n  winner: z.enum(['MODEL_A', 'MODEL_B']),\n  confidence: z.number().min(0).max(1),\n  rationale: z.string().min(1).max(4_000),\n  criteria: z.object({\n    correctness: z.string().min(1).max(1_000),\n    grounding: z.string().min(1).max(1_000),\n    constraintHandling: z.string().min(1).max(1_000),\n    completeness: z.string().min(1).max(1_000),\n  }),\n  violations: z.array(z.string().max(500)).max(20),\n});\nexport type JudgeVerdict = z.infer<typeof JudgeVerdictSchema>;\n\n/**\n * Provider-safe JSON Schema subset generated from the same verdict shape.\n * Runtime Zod parsing above enforces the local string and array limits.\n */\nexport const JUDGE_VERDICT_TRANSPORT_SCHEMA = {\n  type: 'object',\n  additionalProperties: false,\n  required: ['winner', 'confidence', 'rationale', 'criteria', 'violations'],\n  properties: {\n    winner: { type: 'string', enum: ['MODEL_A', 'MODEL_B'] },\n    confidence: { type: 'number', minimum: 0, maximum: 1 },\n    rationale: { type: 'string' },\n    criteria: {\n      type: 'object',\n      additionalProperties: false,\n      required: ['correctness', 'grounding', 'constraintHandling', 'completeness'],\n      properties: {\n        correctness: { type: 'string' },\n        grounding: { type: 'string' },\n        constraintHandling: { type: 'string' },\n        completeness: { type: 'string' },\n      },\n    },\n    violations: { type: 'array', items: { type: 'string' } },\n  },\n} as const;\n\nexport interface JudgeVote {\n  judgeModelId: string;\n  modelAIdentity: string;\n  modelBIdentity: string;\n  verdict: JudgeVerdict | null;\n  winnerModelId: string | null;\n  completion: ModelCompletion | null;\n  error?: string;\n}\n\nexport interface PanelDecision {\n  winnerModelId: string | null;\n  validVotes: number;\n  votesByModel: Record<string, number>;\n  agreement: 'unanimous' | 'split' | 'insufficient';\n  votes: JudgeVote[];\n}\n\nexport interface EloState {\n  ratings: Record<string, number>;\n  points: Record<string, number>;\n}\n\nexport interface MatchResult {\n  methodologyVersion: string;\n  runId: string;\n  /** Absent on legacy journal lines written before run manifests. */\n  runManifestHash?: string;\n  matchId: string;\n  scheduleIndex: number;\n  seed: string;\n  timestamp: string;\n  task: {\n    id: string;\n    version: string;\n    /** Absent on reasoning-only legacy lines. */\n    category?: BenchmarkCategory;\n    cluster: string;\n    publicHash: string;\n    privateHash: string;\n  };\n  competitors: {\n    modelA: string;\n    modelB: string;\n    responseA: CompetitorResponse;\n    responseB: CompetitorResponse;\n  };\n  outcome: 'judged' | 'forfeit' | 'no-contest';\n  winnerModelId: string | null;\n  panel: PanelDecision | null;\n  eloBefore: Record<string, number>;\n  eloAfter: Record<string, number>;\n  pointAwarded: boolean;\n  matchCostUsd: number;\n}\n\nconst NumericRecordSchema = z.record(z.number().finite());\n\nexport const JudgeVoteSchema: z.ZodType<JudgeVote> = z.object({\n  judgeModelId: z.string().min(1),\n  modelAIdentity: z.string().min(1),\n  modelBIdentity: z.string().min(1),\n  verdict: JudgeVerdictSchema.nullable(),\n  winnerModelId: z.string().min(1).nullable(),\n  completion: ModelCompletionSchema.nullable(),\n  error: z.string().optional(),\n});\n\nexport const PanelDecisionSchema: z.ZodType<PanelDecision> = z.object({\n  winnerModelId: z.string().min(1).nullable(),\n  validVotes: z.number().int().nonnegative(),\n  votesByModel: NumericRecordSchema,\n  agreement: z.enum(['unanimous', 'split', 'insufficient']),\n  votes: z.array(JudgeVoteSchema),\n});\n\nexport const MatchResultSchema: z.ZodType<MatchResult> = z.object({\n  methodologyVersion: z.string().min(1),\n  runId: z.string().min(1),\n  runManifestHash: z\n    .string()\n    .regex(/^[a-f0-9]{64}$/)\n    .optional(),\n  matchId: z.string().min(1),\n  scheduleIndex: z.number().int().nonnegative(),\n  seed: z.string(),\n  timestamp: z.string().datetime(),\n  task: z.object({\n    id: z.string().min(1),\n    version: z.string().min(1),\n    category: BenchmarkCategorySchema.optional(),\n    cluster: z.string().min(1),\n    publicHash: z.string().min(1),\n    privateHash: z.string().min(1),\n  }),\n  competitors: z.object({\n    modelA: z.string().min(1),\n    modelB: z.string().min(1),\n    responseA: CompetitorResponseSchema,\n    responseB: CompetitorResponseSchema,\n  }),\n  outcome: z.enum(['judged', 'forfeit', 'no-contest']),\n  winnerModelId: z.string().min(1).nullable(),\n  panel: PanelDecisionSchema.nullable(),\n  eloBefore: NumericRecordSchema,\n  eloAfter: NumericRecordSchema,\n  pointAwarded: z.boolean(),\n  matchCostUsd: z.number().finite().nonnegative(),\n});\n\nexport function competitorCost(response: CompetitorResponse): number {\n  return response.success ? response.costUsd : 0;\n}\n\nexport function competitorOutputTokens(response: CompetitorResponse): number {\n  return response.success ? response.outputTokens : 0;\n}\n\nexport function competitorReasoningTokens(response: CompetitorResponse): number | undefined {\n  return response.success ? response.reasoningTokens : undefined;\n}\n\nexport function competitorContent(response: CompetitorResponse): string {\n  return response.success ? response.content : '';\n}\n","import { z } from 'zod';\n\nexport const ModelRoleSchema = z.enum(['competitor', 'judge']);\nexport type ModelRole = z.infer<typeof ModelRoleSchema>;\n\nexport interface ModelRequestPolicy {\n  maxTokens: number;\n  temperature: number;\n  reasoningEffort: 'high' | 'medium' | 'low';\n  excludeReasoning: boolean;\n  timeoutMs: number;\n}\n\nexport interface ModelRegistryEntry {\n  id: string;\n  canonicalSlug: string;\n  displayName: string;\n  vendor: string;\n  role: ModelRole;\n  enabled: boolean;\n  request: ModelRequestPolicy;\n}\n\nexport const ModelCompletionSchema = z.object({\n  generationId: z.string(),\n  content: z.string(),\n  inputTokens: z.number().finite().nonnegative(),\n  outputTokens: z.number().finite().nonnegative(),\n  reasoningTokens: z.number().finite().nonnegative().optional(),\n  costUsd: z.number().finite().nonnegative(),\n  latencyMs: z.number().finite().nonnegative(),\n  finishReason: z.string(),\n  attempts: z.number().int().positive().optional(),\n});\nexport type ModelCompletion = z.infer<typeof ModelCompletionSchema>;\n\nexport interface ChatRequest {\n  model: ModelRegistryEntry;\n  system: string;\n  user: string;\n  structured?: boolean;\n  /** Execution-only cancellation; never serialized into a run manifest or provider payload. */\n  signal?: AbortSignal;\n  /** Called with the accumulated visible text as it streams in (throttled). */\n  onDelta?: (text: string) => void;\n}\n\nexport interface OpenRouterGateway {\n  complete(request: ChatRequest): Promise<ModelCompletion>;\n  validateModel(model: ModelRegistryEntry, signal?: AbortSignal): Promise<void>;\n}\n","import { z } from 'zod';\n\nimport { BenchmarkCategorySchema } from './categories.js';\n\nexport const TaskArtifactSchema = z\n  .object({\n    id: z.string().min(1).max(80),\n    type: z.enum(['code', 'log', 'config', 'spec', 'diff', 'table', 'note']),\n    label: z.string().min(1).max(160),\n    content: z.string().min(1).max(40_000),\n  })\n  .strict();\nexport type TaskArtifact = z.infer<typeof TaskArtifactSchema>;\n\nexport const TaskPublicSchema = z\n  .object({\n    id: z.string().regex(/^[a-z0-9][a-z0-9-]{2,79}$/),\n    version: z.string().regex(/^\\d+\\.\\d+\\.\\d+$/),\n    category: BenchmarkCategorySchema,\n    cluster: z.string().min(1).max(60),\n    difficulty: z.enum(['hard', 'expert']),\n    title: z.string().min(1).max(180),\n    summary: z.string().min(1).max(500),\n    prompt: z.string().min(1).max(10_000),\n    artifacts: z.array(TaskArtifactSchema).min(1).max(20),\n    tags: z.array(z.string().min(1).max(60)).default([]),\n  })\n  .strict();\nexport type TaskPublic = z.infer<typeof TaskPublicSchema>;\n\nexport const TaskPrivateSchema = z\n  .object({\n    id: z.string().min(1),\n    version: z.string().min(1),\n    expectedResolution: z.string().min(1).max(10_000),\n    requiredEvidence: z.array(z.string().min(1)).min(1),\n    disqualifyingErrors: z.array(z.string().min(1)).default([]),\n    rubric: z\n      .object({\n        correctness: z.string().min(1),\n        evidenceGrounding: z.string().min(1),\n        constraintHandling: z.string().min(1),\n        completeness: z.string().min(1),\n      })\n      .strict(),\n  })\n  .strict();\nexport type TaskPrivate = z.infer<typeof TaskPrivateSchema>;\n\nexport interface ArenaTask {\n  public: TaskPublic;\n  /** Null when loaded without a private overlay. */\n  private: TaskPrivate | null;\n  publicHash: string;\n  privateHash: string | null;\n}\n\n/** A task whose hidden reference is present — required for judged matches. */\nexport type CompleteArenaTask = ArenaTask & {\n  private: TaskPrivate;\n  privateHash: string;\n};\n","/**\n * Public BridgeBench arena contract version.\n *\n * Increment this when an exported contract changes incompatibly. The entry\n * point intentionally exports contracts only; it has no arena-runner or\n * provider implementation dependency.\n */\nexport const CONTRACTS_VERSION = '1.0.0' as const;\n\nexport * from './categories.js';\nexport * from './events.js';\nexport * from './journal.js';\nexport * from './models.js';\nexport * from './reports.js';\nexport * from './tasks.js';\n"]}